RES — Model quality is non-monotonic with size within families
Model quality is non-monotonic with size within families: the 3B model beats the 7B on triage F1-macro, and two of the largest models fall below several smaller ones, showing parameter scaling does not guarantee better triage performance.
Source: PUMA project documentation · Traceability: corpus unit
MEM-043· Confidence: verified-at-primary-source